AI · Nov 18, 2026
Gemini comes to budget Android phones, and that is a bigger deal than it soundsSignal turns off screenshots to keep OS-level AI agents out of private chats
· Updated · Signal Desk newsroom
Analysis of reporting by The Verge. How we work

Signal now blocks screenshots by default on iOS 27 and Android so Siri, Gemini and other assistants can’t read its conversations. The fix is admittedly partial, and the real target is Apple, Google and Microsoft.
Key takeaways
- Signal is restricting screenshots in its iOS 27 and Android apps so on-device AI agents can’t read message content on screen.
- Users can switch screenshots back on, but Signal warns that doing so can expose their and their contacts’ messages to those agents.
- Signal president Meredith Whittaker calls the fix “incomplete”: agents could still reach app data through other paths such as the file system, and there is no macOS version yet.
- Signal is asking Apple, Google and Microsoft for simple developer controls and accurate documentation on how agents access app data.
- Third-party agents such as Meta’s Muse that lean on accessibility permissions are named as a separate threat.
A privacy app pulls the plug on its own feature
Signal is giving up a small convenience, screenshots, to protect a big promise. The encrypted messenger now restricts screenshots in its Android and iOS 27 apps. The point isn’t to stop people from saving chats. It’s to stop the AI assistants built into phone operating systems, including Siri and Gemini, from reading whatever is on screen. According to The Verge, Signal president Meredith Whittaker said in a blog post that these OS-level agents are undermining Signal’s ability to guarantee privacy on its own.
People can restore screenshots in settings. Signal warns that doing so could expose their messages, and their contacts’ messages, to those agents. That second part matters. In a group chat, one member’s settings decide how private everyone else’s words are. That was always true of screenshots. It hits harder when the thing taking the picture is an assistant that summarizes, stores and acts on what it sees.
Why this is bigger than a toggle
End-to-end encryption protects messages in transit and on Signal’s servers. It does nothing once the decrypted text is on a screen that the operating system is allowed to inspect. For years that gap was mostly theoretical: malware, a nosy shoulder, a screenshot sent onward. Agentic assistants make it routine. An OS that reads the screen to “learn more about the user” treats every app’s private view as raw material.
Signal has done this before. Last year it used a similar screen-capture block to keep Microsoft’s Recall from recording secure chats. The difference now is reach. This isn’t one optional Windows feature anymore; it’s the default direction of both major phone platforms. Recent Signal Desk coverage shows how fast the agent surface is growing, from Google letting any AI agent run your smart home to websites closing their doors to AI agents. Signal is the messaging version of that pushback.
The honest part: it doesn’t fully work
We give Signal credit for saying plainly that the fix is “incomplete.” Blocking screenshots shuts one door. Agents could still reach app data through the device file system, and Signal says it can’t even ship the protection on macOS. The reason is the core of its complaint. Apple, Google and Microsoft don’t give developers a clean way to say “agents may not read this view” without breaking the normal experience.
Signal also points to third-party agents, naming Meta’s Muse, that use deep accessibility permissions like a skeleton key. Accessibility APIs exist so screen readers and assistive tools can work across every app. That makes them hard to lock down without hurting disabled users. Signal is right that platforms need finer controls. Those controls also have to keep real accessibility tools working, which is a tougher design problem than the post lets on.
Who this lands on
- Signal users
- Better default protection against assistant snooping, at the cost of screenshots. Group privacy still depends on every member keeping the setting on.
- Apple, Google, Microsoft
- Public pressure to ship per-app agent-access controls and documentation. Ignoring it hands privacy-focused rivals a clear story.
- Other app developers
- Banking, health and enterprise apps face the same exposure and lack the same tools. Expect copycat screenshot blocks as a stopgap.
- Agent makers
- Every opt-out like this shrinks what screen-reading assistants can see, which pushes them toward formal, permissioned integrations.
What to watch
- Whether Apple or Google answers with a developer flag that keeps agents out of sensitive views without disabling screenshots altogether.
- A macOS version of the protection, which Signal says it currently can’t build.
- Other secure-messaging and finance apps adopting similar blocks.
- How Meta and other third-party agent makers respond to the accessibility “skeleton key” criticism.
- Any platform policy change limiting agent use of accessibility services, and how it treats assistive-technology users.
FAQ
Does this stop all AI access to Signal messages?
No. Signal itself calls it incomplete. It blocks screen capture, but agents could still reach app data through other routes, such as the file system.
Can I still take screenshots in Signal?
Yes. You can turn screenshots back on in settings, but Signal warns that this may expose your messages and your contacts’ messages to on-device AI agents.
Why isn’t this on Mac?
Signal says the platform tools don’t let it restrict AI data access on macOS without drastic compromises to the user experience.
What is Signal asking platforms to do?
Provide simple technical tools and accurate documentation so developers can control what AI agents read inside their apps.
Companies: Google
More on Google
AI · Nov 10, 2026
Gemini 4 Carbon opens to developers at Argon’s price until November 30, then doublesAI · Nov 10, 2026
Gemini for healthcare targets prior authorization, the paperwork where patients have the most at stakeAI · Nov 12, 2026
ChatGPT, Gemini and Perplexity sent shoppers to fake stores in a 6,795-answer testAI · Oct 30, 2026
A single robots.txt line to block AI training: what the OpenAI-Google-Microsoft-Anthropic pledge does and doesn’t fix